xref: /linux/arch/powerpc/mm/mem.c (revision 26deb04342e343ac58ab05bc7d2345ff0be9b667) 
1 /*
2  *  PowerPC version
3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4  *
5  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
6  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
7  *    Copyright (C) 1996 Paul Mackerras
8  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
9  *
10  *  Derived from "arch/i386/mm/init.c"
11  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
12  *
13  *  This program is free software; you can redistribute it and/or
14  *  modify it under the terms of the GNU General Public License
15  *  as published by the Free Software Foundation; either version
16  *  2 of the License, or (at your option) any later version.
17  *
18  */
19 
20 #include <linux/export.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/string.h>
25 #include <linux/gfp.h>
26 #include <linux/types.h>
27 #include <linux/mm.h>
28 #include <linux/stddef.h>
29 #include <linux/init.h>
30 #include <linux/memblock.h>
31 #include <linux/highmem.h>
32 #include <linux/initrd.h>
33 #include <linux/pagemap.h>
34 #include <linux/suspend.h>
35 #include <linux/hugetlb.h>
36 #include <linux/slab.h>
37 #include <linux/vmalloc.h>
38 #include <linux/memremap.h>
39 
40 #include <asm/pgalloc.h>
41 #include <asm/prom.h>
42 #include <asm/io.h>
43 #include <asm/mmu_context.h>
44 #include <asm/pgtable.h>
45 #include <asm/mmu.h>
46 #include <asm/smp.h>
47 #include <asm/machdep.h>
48 #include <asm/btext.h>
49 #include <asm/tlb.h>
50 #include <asm/sections.h>
51 #include <asm/sparsemem.h>
52 #include <asm/vdso.h>
53 #include <asm/fixmap.h>
54 #include <asm/swiotlb.h>
55 #include <asm/rtas.h>
56 
57 #include <mm/mmu_decl.h>
58 
59 #ifndef CPU_FTR_COHERENT_ICACHE
60 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
61 #define CPU_FTR_NOEXECUTE	0
62 #endif
63 
64 unsigned long long memory_limit;
65 bool init_mem_is_free;
66 
67 #ifdef CONFIG_HIGHMEM
68 pte_t *kmap_pte;
69 EXPORT_SYMBOL(kmap_pte);
70 pgprot_t kmap_prot;
71 EXPORT_SYMBOL(kmap_prot);
72 
73 static inline pte_t *virt_to_kpte(unsigned long vaddr)
74 {
75 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
76 			vaddr), vaddr), vaddr);
77 }
78 #endif
79 
80 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
81 			      unsigned long size, pgprot_t vma_prot)
82 {
83 	if (ppc_md.phys_mem_access_prot)
84 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
85 
86 	if (!page_is_ram(pfn))
87 		vma_prot = pgprot_noncached(vma_prot);
88 
89 	return vma_prot;
90 }
91 EXPORT_SYMBOL(phys_mem_access_prot);
92 
93 #ifdef CONFIG_MEMORY_HOTPLUG
94 
95 #ifdef CONFIG_NUMA
96 int memory_add_physaddr_to_nid(u64 start)
97 {
98 	return hot_add_scn_to_nid(start);
99 }
100 #endif
101 
102 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
103 {
104 	return -ENODEV;
105 }
106 
107 int __weak remove_section_mapping(unsigned long start, unsigned long end)
108 {
109 	return -ENODEV;
110 }
111 
112 int __ref arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
113 			  bool want_memblock)
114 {
115 	unsigned long start_pfn = start >> PAGE_SHIFT;
116 	unsigned long nr_pages = size >> PAGE_SHIFT;
117 	int rc;
118 
119 	resize_hpt_for_hotplug(memblock_phys_mem_size());
120 
121 	start = (unsigned long)__va(start);
122 	rc = create_section_mapping(start, start + size, nid);
123 	if (rc) {
124 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
125 			start, start + size, rc);
126 		return -EFAULT;
127 	}
128 	flush_inval_dcache_range(start, start + size);
129 
130 	return __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
131 }
132 
133 #ifdef CONFIG_MEMORY_HOTREMOVE
134 int __ref arch_remove_memory(int nid, u64 start, u64 size,
135 			     struct vmem_altmap *altmap)
136 {
137 	unsigned long start_pfn = start >> PAGE_SHIFT;
138 	unsigned long nr_pages = size >> PAGE_SHIFT;
139 	struct page *page;
140 	int ret;
141 
142 	/*
143 	 * If we have an altmap then we need to skip over any reserved PFNs
144 	 * when querying the zone.
145 	 */
146 	page = pfn_to_page(start_pfn);
147 	if (altmap)
148 		page += vmem_altmap_offset(altmap);
149 
150 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
151 	if (ret)
152 		return ret;
153 
154 	/* Remove htab bolted mappings for this section of memory */
155 	start = (unsigned long)__va(start);
156 	flush_inval_dcache_range(start, start + size);
157 	ret = remove_section_mapping(start, start + size);
158 
159 	/* Ensure all vmalloc mappings are flushed in case they also
160 	 * hit that section of memory
161 	 */
162 	vm_unmap_aliases();
163 
164 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
165 		pr_warn("Hash collision while resizing HPT\n");
166 
167 	return ret;
168 }
169 #endif
170 #endif /* CONFIG_MEMORY_HOTPLUG */
171 
172 #ifndef CONFIG_NEED_MULTIPLE_NODES
173 void __init mem_topology_setup(void)
174 {
175 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
176 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
177 #ifdef CONFIG_HIGHMEM
178 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
179 #endif
180 
181 	/* Place all memblock_regions in the same node and merge contiguous
182 	 * memblock_regions
183 	 */
184 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
185 }
186 
187 void __init initmem_init(void)
188 {
189 	/* XXX need to clip this if using highmem? */
190 	sparse_memory_present_with_active_regions(0);
191 	sparse_init();
192 }
193 
194 /* mark pages that don't exist as nosave */
195 static int __init mark_nonram_nosave(void)
196 {
197 	struct memblock_region *reg, *prev = NULL;
198 
199 	for_each_memblock(memory, reg) {
200 		if (prev &&
201 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
202 			register_nosave_region(memblock_region_memory_end_pfn(prev),
203 					       memblock_region_memory_base_pfn(reg));
204 		prev = reg;
205 	}
206 	return 0;
207 }
208 #else /* CONFIG_NEED_MULTIPLE_NODES */
209 static int __init mark_nonram_nosave(void)
210 {
211 	return 0;
212 }
213 #endif
214 
215 /*
216  * Zones usage:
217  *
218  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
219  * everything else. GFP_DMA32 page allocations automatically fall back to
220  * ZONE_DMA.
221  *
222  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
223  * inform the generic DMA mapping code.  32-bit only devices (if not handled
224  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
225  * otherwise served by ZONE_DMA.
226  */
227 static unsigned long max_zone_pfns[MAX_NR_ZONES];
228 
229 /*
230  * paging_init() sets up the page tables - in fact we've already done this.
231  */
232 void __init paging_init(void)
233 {
234 	unsigned long long total_ram = memblock_phys_mem_size();
235 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
236 
237 #ifdef CONFIG_PPC32
238 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
239 	unsigned long end = __fix_to_virt(FIX_HOLE);
240 
241 	for (; v < end; v += PAGE_SIZE)
242 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
243 #endif
244 
245 #ifdef CONFIG_HIGHMEM
246 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
247 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
248 
249 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
250 	kmap_prot = PAGE_KERNEL;
251 #endif /* CONFIG_HIGHMEM */
252 
253 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
254 	       (unsigned long long)top_of_ram, total_ram);
255 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
256 	       (long int)((top_of_ram - total_ram) >> 20));
257 
258 #ifdef CONFIG_ZONE_DMA
259 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
260 #endif
261 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
262 #ifdef CONFIG_HIGHMEM
263 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
264 #endif
265 
266 	free_area_init_nodes(max_zone_pfns);
267 
268 	mark_nonram_nosave();
269 }
270 
271 void __init mem_init(void)
272 {
273 	/*
274 	 * book3s is limited to 16 page sizes due to encoding this in
275 	 * a 4-bit field for slices.
276 	 */
277 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
278 
279 #ifdef CONFIG_SWIOTLB
280 	swiotlb_init(0);
281 #endif
282 
283 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
284 	set_max_mapnr(max_pfn);
285 	memblock_free_all();
286 
287 #ifdef CONFIG_HIGHMEM
288 	{
289 		unsigned long pfn, highmem_mapnr;
290 
291 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
292 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
293 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
294 			struct page *page = pfn_to_page(pfn);
295 			if (!memblock_is_reserved(paddr))
296 				free_highmem_page(page);
297 		}
298 	}
299 #endif /* CONFIG_HIGHMEM */
300 
301 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
302 	/*
303 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
304 	 * functions.... do it here for the non-smp case.
305 	 */
306 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
307 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
308 #endif
309 
310 	mem_init_print_info(NULL);
311 #ifdef CONFIG_PPC32
312 	pr_info("Kernel virtual memory layout:\n");
313 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
314 #ifdef CONFIG_HIGHMEM
315 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
316 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
317 #endif /* CONFIG_HIGHMEM */
318 #ifdef CONFIG_NOT_COHERENT_CACHE
319 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
320 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
321 #endif /* CONFIG_NOT_COHERENT_CACHE */
322 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
323 		ioremap_bot, IOREMAP_TOP);
324 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
325 		VMALLOC_START, VMALLOC_END);
326 #endif /* CONFIG_PPC32 */
327 }
328 
329 void free_initmem(void)
330 {
331 	ppc_md.progress = ppc_printk_progress;
332 	mark_initmem_nx();
333 	init_mem_is_free = true;
334 	free_initmem_default(POISON_FREE_INITMEM);
335 }
336 
337 #ifdef CONFIG_BLK_DEV_INITRD
338 void __init free_initrd_mem(unsigned long start, unsigned long end)
339 {
340 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
341 }
342 #endif
343 
344 /*
345  * This is called when a page has been modified by the kernel.
346  * It just marks the page as not i-cache clean.  We do the i-cache
347  * flush later when the page is given to a user process, if necessary.
348  */
349 void flush_dcache_page(struct page *page)
350 {
351 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
352 		return;
353 	/* avoid an atomic op if possible */
354 	if (test_bit(PG_arch_1, &page->flags))
355 		clear_bit(PG_arch_1, &page->flags);
356 }
357 EXPORT_SYMBOL(flush_dcache_page);
358 
359 void flush_dcache_icache_page(struct page *page)
360 {
361 #ifdef CONFIG_HUGETLB_PAGE
362 	if (PageCompound(page)) {
363 		flush_dcache_icache_hugepage(page);
364 		return;
365 	}
366 #endif
367 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
368 	/* On 8xx there is no need to kmap since highmem is not supported */
369 	__flush_dcache_icache(page_address(page));
370 #else
371 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
372 		void *start = kmap_atomic(page);
373 		__flush_dcache_icache(start);
374 		kunmap_atomic(start);
375 	} else {
376 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
377 	}
378 #endif
379 }
380 EXPORT_SYMBOL(flush_dcache_icache_page);
381 
382 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
383 {
384 	clear_page(page);
385 
386 	/*
387 	 * We shouldn't have to do this, but some versions of glibc
388 	 * require it (ld.so assumes zero filled pages are icache clean)
389 	 * - Anton
390 	 */
391 	flush_dcache_page(pg);
392 }
393 EXPORT_SYMBOL(clear_user_page);
394 
395 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
396 		    struct page *pg)
397 {
398 	copy_page(vto, vfrom);
399 
400 	/*
401 	 * We should be able to use the following optimisation, however
402 	 * there are two problems.
403 	 * Firstly a bug in some versions of binutils meant PLT sections
404 	 * were not marked executable.
405 	 * Secondly the first word in the GOT section is blrl, used
406 	 * to establish the GOT address. Until recently the GOT was
407 	 * not marked executable.
408 	 * - Anton
409 	 */
410 #if 0
411 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
412 		return;
413 #endif
414 
415 	flush_dcache_page(pg);
416 }
417 
418 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
419 			     unsigned long addr, int len)
420 {
421 	unsigned long maddr;
422 
423 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
424 	flush_icache_range(maddr, maddr + len);
425 	kunmap(page);
426 }
427 EXPORT_SYMBOL(flush_icache_user_range);
428 
429 /*
430  * This is called at the end of handling a user page fault, when the
431  * fault has been handled by updating a PTE in the linux page tables.
432  * We use it to preload an HPTE into the hash table corresponding to
433  * the updated linux PTE.
434  *
435  * This must always be called with the pte lock held.
436  */
437 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
438 		      pte_t *ptep)
439 {
440 #ifdef CONFIG_PPC_BOOK3S
441 	/*
442 	 * We don't need to worry about _PAGE_PRESENT here because we are
443 	 * called with either mm->page_table_lock held or ptl lock held
444 	 */
445 	unsigned long trap;
446 	bool is_exec;
447 
448 	if (radix_enabled()) {
449 		prefetch((void *)address);
450 		return;
451 	}
452 
453 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
454 	if (!pte_young(*ptep) || address >= TASK_SIZE)
455 		return;
456 
457 	/* We try to figure out if we are coming from an instruction
458 	 * access fault and pass that down to __hash_page so we avoid
459 	 * double-faulting on execution of fresh text. We have to test
460 	 * for regs NULL since init will get here first thing at boot
461 	 *
462 	 * We also avoid filling the hash if not coming from a fault
463 	 */
464 
465 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
466 	switch (trap) {
467 	case 0x300:
468 		is_exec = false;
469 		break;
470 	case 0x400:
471 		is_exec = true;
472 		break;
473 	default:
474 		return;
475 	}
476 
477 	hash_preload(vma->vm_mm, address, is_exec, trap);
478 #endif /* CONFIG_PPC_BOOK3S */
479 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
480 	&& defined(CONFIG_HUGETLB_PAGE)
481 	if (is_vm_hugetlb_page(vma))
482 		book3e_hugetlb_preload(vma, address, *ptep);
483 #endif
484 }
485 
486 /*
487  * System memory should not be in /proc/iomem but various tools expect it
488  * (eg kdump).
489  */
490 static int __init add_system_ram_resources(void)
491 {
492 	struct memblock_region *reg;
493 
494 	for_each_memblock(memory, reg) {
495 		struct resource *res;
496 		unsigned long base = reg->base;
497 		unsigned long size = reg->size;
498 
499 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
500 		WARN_ON(!res);
501 
502 		if (res) {
503 			res->name = "System RAM";
504 			res->start = base;
505 			res->end = base + size - 1;
506 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
507 			WARN_ON(request_resource(&iomem_resource, res) < 0);
508 		}
509 	}
510 
511 	return 0;
512 }
513 subsys_initcall(add_system_ram_resources);
514 
515 #ifdef CONFIG_STRICT_DEVMEM
516 /*
517  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
518  * is valid. The argument is a physical page number.
519  *
520  * Access has to be given to non-kernel-ram areas as well, these contain the
521  * PCI mmio resources as well as potential bios/acpi data regions.
522  */
523 int devmem_is_allowed(unsigned long pfn)
524 {
525 	if (page_is_rtas_user_buf(pfn))
526 		return 1;
527 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
528 		return 0;
529 	if (!page_is_ram(pfn))
530 		return 1;
531 	return 0;
532 }
533 #endif /* CONFIG_STRICT_DEVMEM */
534 
535 /*
536  * This is defined in kernel/resource.c but only powerpc needs to export it, for
537  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
538  */
539 EXPORT_SYMBOL_GPL(walk_system_ram_range);
540